xref: /freebsd/sys/kern/uipc_usrreq.c (revision cacdd70cc751fb68dec4b86c5e5b8c969b6e26ef)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2004-2007 Robert N. M. Watson
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 4. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	From: @(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
32  */
33 
34 /*
35  * UNIX Domain (Local) Sockets
36  *
37  * This is an implementation of UNIX (local) domain sockets.  Each socket has
38  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
39  * may be connected to 0 or 1 other socket.  Datagram sockets may be
40  * connected to 0, 1, or many other sockets.  Sockets may be created and
41  * connected in pairs (socketpair(2)), or bound/connected to using the file
42  * system name space.  For most purposes, only the receive socket buffer is
43  * used, as sending on one socket delivers directly to the receive socket
44  * buffer of a second socket.
45  *
46  * The implementation is substantially complicated by the fact that
47  * "ancillary data", such as file descriptors or credentials, may be passed
48  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
49  * over other UNIX domain sockets requires the implementation of a simple
50  * garbage collector to find and tear down cycles of disconnected sockets.
51  *
52  * TODO:
53  *	SEQPACKET, RDM
54  *	rethink name space problems
55  *	need a proper out-of-band
56  */
57 
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
60 
61 #include "opt_ddb.h"
62 #include "opt_mac.h"
63 
64 #include <sys/param.h>
65 #include <sys/domain.h>
66 #include <sys/fcntl.h>
67 #include <sys/malloc.h>		/* XXX must be before <sys/file.h> */
68 #include <sys/eventhandler.h>
69 #include <sys/file.h>
70 #include <sys/filedesc.h>
71 #include <sys/jail.h>
72 #include <sys/kernel.h>
73 #include <sys/lock.h>
74 #include <sys/mbuf.h>
75 #include <sys/mount.h>
76 #include <sys/mutex.h>
77 #include <sys/namei.h>
78 #include <sys/proc.h>
79 #include <sys/protosw.h>
80 #include <sys/resourcevar.h>
81 #include <sys/rwlock.h>
82 #include <sys/socket.h>
83 #include <sys/socketvar.h>
84 #include <sys/signalvar.h>
85 #include <sys/stat.h>
86 #include <sys/sx.h>
87 #include <sys/sysctl.h>
88 #include <sys/systm.h>
89 #include <sys/taskqueue.h>
90 #include <sys/un.h>
91 #include <sys/unpcb.h>
92 #include <sys/vnode.h>
93 
94 #ifdef DDB
95 #include <ddb/ddb.h>
96 #endif
97 
98 #include <security/mac/mac_framework.h>
99 
100 #include <vm/uma.h>
101 
102 static uma_zone_t	unp_zone;
103 static unp_gen_t	unp_gencnt;
104 static u_int		unp_count;	/* Count of local sockets. */
105 static ino_t		unp_ino;	/* Prototype for fake inode numbers. */
106 static int		unp_rights;	/* File descriptors in flight. */
107 static struct unp_head	unp_shead;	/* List of local stream sockets. */
108 static struct unp_head	unp_dhead;	/* List of local datagram sockets. */
109 
110 static const struct sockaddr	sun_noname = { sizeof(sun_noname), AF_LOCAL };
111 
112 /*
113  * Garbage collection of cyclic file descriptor/socket references occurs
114  * asynchronously in a taskqueue context in order to avoid recursion and
115  * reentrance in the UNIX domain socket, file descriptor, and socket layer
116  * code.  See unp_gc() for a full description.
117  */
118 static struct task	unp_gc_task;
119 
120 /*
121  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
122  * stream sockets, although the total for sender and receiver is actually
123  * only PIPSIZ.
124  *
125  * Datagram sockets really use the sendspace as the maximum datagram size,
126  * and don't really want to reserve the sendspace.  Their recvspace should be
127  * large enough for at least one max-size datagram plus address.
128  */
129 #ifndef PIPSIZ
130 #define	PIPSIZ	8192
131 #endif
132 static u_long	unpst_sendspace = PIPSIZ;
133 static u_long	unpst_recvspace = PIPSIZ;
134 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
135 static u_long	unpdg_recvspace = 4*1024;
136 
137 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
138 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
139 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
140 
141 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
142 	   &unpst_sendspace, 0, "Default stream send space.");
143 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
144 	   &unpst_recvspace, 0, "Default stream receive space.");
145 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
146 	   &unpdg_sendspace, 0, "Default datagram send space.");
147 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
148 	   &unpdg_recvspace, 0, "Default datagram receive space.");
149 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
150     "File descriptors in flight.");
151 
152 /*-
153  * Locking and synchronization:
154  *
155  * The global UNIX domain socket rwlock (unp_global_rwlock) protects all
156  * global variables, including the linked lists tracking the set of allocated
157  * UNIX domain sockets.  The global rwlock also serves to prevent deadlock
158  * when more than one PCB lock is acquired at a time (i.e., during
159  * connect()).  Finally, the global rwlock protects uncounted references from
160  * vnodes to sockets bound to those vnodes: to safely dereference the
161  * v_socket pointer, the global rwlock must be held while a full reference is
162  * acquired.
163  *
164  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
165  * allocated in pru_attach() and freed in pru_detach().  The validity of that
166  * pointer is an invariant, so no lock is required to dereference the so_pcb
167  * pointer if a valid socket reference is held by the caller.  In practice,
168  * this is always true during operations performed on a socket.  Each unpcb
169  * has a back-pointer to its socket, unp_socket, which will be stable under
170  * the same circumstances.
171  *
172  * This pointer may only be safely dereferenced as long as a valid reference
173  * to the unpcb is held.  Typically, this reference will be from the socket,
174  * or from another unpcb when the referring unpcb's lock is held (in order
175  * that the reference not be invalidated during use).  For example, to follow
176  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
177  * as unp_socket remains valid as long as the reference to unp_conn is valid.
178  *
179  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
180  * atomic reads without the lock may be performed "lockless", but more
181  * complex reads and read-modify-writes require the mutex to be held.  No
182  * lock order is defined between unpcb locks -- multiple unpcb locks may be
183  * acquired at the same time only when holding the global UNIX domain socket
184  * rwlock exclusively, which prevents deadlocks.
185  *
186  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
187  * protocols, bind() is a non-atomic operation, and connect() requires
188  * potential sleeping in the protocol, due to potentially waiting on local or
189  * distributed file systems.  We try to separate "lookup" operations, which
190  * may sleep, and the IPC operations themselves, which typically can occur
191  * with relative atomicity as locks can be held over the entire operation.
192  *
193  * Another tricky issue is simultaneous multi-threaded or multi-process
194  * access to a single UNIX domain socket.  These are handled by the flags
195  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
196  * binding, both of which involve dropping UNIX domain socket locks in order
197  * to perform namei() and other file system operations.
198  */
199 static struct rwlock	unp_global_rwlock;
200 
201 #define	UNP_GLOBAL_LOCK_INIT()		rw_init(&unp_global_rwlock,	\
202 					    "unp_global_rwlock")
203 
204 #define	UNP_GLOBAL_LOCK_ASSERT()	rw_assert(&unp_global_rwlock,	\
205 					    RA_LOCKED)
206 #define	UNP_GLOBAL_UNLOCK_ASSERT()	rw_assert(&unp_global_rwlock,	\
207 					    RA_UNLOCKED)
208 
209 #define	UNP_GLOBAL_WLOCK()		rw_wlock(&unp_global_rwlock)
210 #define	UNP_GLOBAL_WUNLOCK()		rw_wunlock(&unp_global_rwlock)
211 #define	UNP_GLOBAL_WLOCK_ASSERT()	rw_assert(&unp_global_rwlock,	\
212 					    RA_WLOCKED)
213 #define	UNP_GLOBAL_WOWNED()		rw_wowned(&unp_global_rwlock)
214 
215 #define	UNP_GLOBAL_RLOCK()		rw_rlock(&unp_global_rwlock)
216 #define	UNP_GLOBAL_RUNLOCK()		rw_runlock(&unp_global_rwlock)
217 #define	UNP_GLOBAL_RLOCK_ASSERT()	rw_assert(&unp_global_rwlock,	\
218 					    RA_RLOCKED)
219 
220 #define UNP_PCB_LOCK_INIT(unp)		mtx_init(&(unp)->unp_mtx,	\
221 					    "unp_mtx", "unp_mtx",	\
222 					    MTX_DUPOK|MTX_DEF|MTX_RECURSE)
223 #define	UNP_PCB_LOCK_DESTROY(unp)	mtx_destroy(&(unp)->unp_mtx)
224 #define	UNP_PCB_LOCK(unp)		mtx_lock(&(unp)->unp_mtx)
225 #define	UNP_PCB_UNLOCK(unp)		mtx_unlock(&(unp)->unp_mtx)
226 #define	UNP_PCB_LOCK_ASSERT(unp)	mtx_assert(&(unp)->unp_mtx, MA_OWNED)
227 
228 static int	unp_connect(struct socket *, struct sockaddr *,
229 		    struct thread *);
230 static int	unp_connect2(struct socket *so, struct socket *so2, int);
231 static void	unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
232 static void	unp_shutdown(struct unpcb *);
233 static void	unp_drop(struct unpcb *, int);
234 static void	unp_gc(__unused void *, int);
235 static void	unp_scan(struct mbuf *, void (*)(struct file *));
236 static void	unp_discard(struct file *);
237 static void	unp_freerights(struct file **, int);
238 static int	unp_internalize(struct mbuf **, struct thread *);
239 static void	unp_internalize_fp(struct file *);
240 static void	unp_externalize_fp(struct file *);
241 static struct mbuf	*unp_addsockcred(struct thread *, struct mbuf *);
242 
243 /*
244  * Definitions of protocols supported in the LOCAL domain.
245  */
246 static struct domain localdomain;
247 static struct protosw localsw[] = {
248 {
249 	.pr_type =		SOCK_STREAM,
250 	.pr_domain =		&localdomain,
251 	.pr_flags =		PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
252 	.pr_ctloutput =		&uipc_ctloutput,
253 	.pr_usrreqs =		&uipc_usrreqs
254 },
255 {
256 	.pr_type =		SOCK_DGRAM,
257 	.pr_domain =		&localdomain,
258 	.pr_flags =		PR_ATOMIC|PR_ADDR|PR_RIGHTS,
259 	.pr_usrreqs =		&uipc_usrreqs
260 },
261 };
262 
263 static struct domain localdomain = {
264 	.dom_family =		AF_LOCAL,
265 	.dom_name =		"local",
266 	.dom_init =		unp_init,
267 	.dom_externalize =	unp_externalize,
268 	.dom_dispose =		unp_dispose,
269 	.dom_protosw =		localsw,
270 	.dom_protoswNPROTOSW =	&localsw[sizeof(localsw)/sizeof(localsw[0])]
271 };
272 DOMAIN_SET(local);
273 
274 static void
275 uipc_abort(struct socket *so)
276 {
277 	struct unpcb *unp, *unp2;
278 
279 	unp = sotounpcb(so);
280 	KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
281 
282 	UNP_GLOBAL_WLOCK();
283 	UNP_PCB_LOCK(unp);
284 	unp2 = unp->unp_conn;
285 	if (unp2 != NULL) {
286 		UNP_PCB_LOCK(unp2);
287 		unp_drop(unp2, ECONNABORTED);
288 		UNP_PCB_UNLOCK(unp2);
289 	}
290 	UNP_PCB_UNLOCK(unp);
291 	UNP_GLOBAL_WUNLOCK();
292 }
293 
294 static int
295 uipc_accept(struct socket *so, struct sockaddr **nam)
296 {
297 	struct unpcb *unp, *unp2;
298 	const struct sockaddr *sa;
299 
300 	/*
301 	 * Pass back name of connected socket, if it was bound and we are
302 	 * still connected (our peer may have closed already!).
303 	 */
304 	unp = sotounpcb(so);
305 	KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
306 
307 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
308 	UNP_GLOBAL_RLOCK();
309 	unp2 = unp->unp_conn;
310 	if (unp2 != NULL && unp2->unp_addr != NULL) {
311 		UNP_PCB_LOCK(unp2);
312 		sa = (struct sockaddr *) unp2->unp_addr;
313 		bcopy(sa, *nam, sa->sa_len);
314 		UNP_PCB_UNLOCK(unp2);
315 	} else {
316 		sa = &sun_noname;
317 		bcopy(sa, *nam, sa->sa_len);
318 	}
319 	UNP_GLOBAL_RUNLOCK();
320 	return (0);
321 }
322 
323 static int
324 uipc_attach(struct socket *so, int proto, struct thread *td)
325 {
326 	u_long sendspace, recvspace;
327 	struct unpcb *unp;
328 	int error, locked;
329 
330 	KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
331 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
332 		switch (so->so_type) {
333 		case SOCK_STREAM:
334 			sendspace = unpst_sendspace;
335 			recvspace = unpst_recvspace;
336 			break;
337 
338 		case SOCK_DGRAM:
339 			sendspace = unpdg_sendspace;
340 			recvspace = unpdg_recvspace;
341 			break;
342 
343 		default:
344 			panic("uipc_attach");
345 		}
346 		error = soreserve(so, sendspace, recvspace);
347 		if (error)
348 			return (error);
349 	}
350 	unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
351 	if (unp == NULL)
352 		return (ENOBUFS);
353 	LIST_INIT(&unp->unp_refs);
354 	UNP_PCB_LOCK_INIT(unp);
355 	unp->unp_socket = so;
356 	so->so_pcb = unp;
357 	unp->unp_refcount = 1;
358 
359 	/*
360 	 * uipc_attach() may be called indirectly from within the UNIX domain
361 	 * socket code via sonewconn() in unp_connect().  Since rwlocks can
362 	 * not be recursed, we do the closest thing.
363 	 */
364 	locked = 0;
365 	if (!UNP_GLOBAL_WOWNED()) {
366 		UNP_GLOBAL_WLOCK();
367 		locked = 1;
368 	}
369 	unp->unp_gencnt = ++unp_gencnt;
370 	unp_count++;
371 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead,
372 	    unp, unp_link);
373 	if (locked)
374 		UNP_GLOBAL_WUNLOCK();
375 
376 	return (0);
377 }
378 
379 static int
380 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
381 {
382 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
383 	struct vattr vattr;
384 	int error, namelen, vfslocked;
385 	struct nameidata nd;
386 	struct unpcb *unp;
387 	struct vnode *vp;
388 	struct mount *mp;
389 	char *buf;
390 
391 	unp = sotounpcb(so);
392 	KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
393 
394 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
395 	if (namelen <= 0)
396 		return (EINVAL);
397 
398 	/*
399 	 * We don't allow simultaneous bind() calls on a single UNIX domain
400 	 * socket, so flag in-progress operations, and return an error if an
401 	 * operation is already in progress.
402 	 *
403 	 * Historically, we have not allowed a socket to be rebound, so this
404 	 * also returns an error.  Not allowing re-binding simplifies the
405 	 * implementation and avoids a great many possible failure modes.
406 	 */
407 	UNP_PCB_LOCK(unp);
408 	if (unp->unp_vnode != NULL) {
409 		UNP_PCB_UNLOCK(unp);
410 		return (EINVAL);
411 	}
412 	if (unp->unp_flags & UNP_BINDING) {
413 		UNP_PCB_UNLOCK(unp);
414 		return (EALREADY);
415 	}
416 	unp->unp_flags |= UNP_BINDING;
417 	UNP_PCB_UNLOCK(unp);
418 
419 	buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
420 	bcopy(soun->sun_path, buf, namelen);
421 	buf[namelen] = 0;
422 
423 restart:
424 	vfslocked = 0;
425 	NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME,
426 	    UIO_SYSSPACE, buf, td);
427 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
428 	error = namei(&nd);
429 	if (error)
430 		goto error;
431 	vp = nd.ni_vp;
432 	vfslocked = NDHASGIANT(&nd);
433 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
434 		NDFREE(&nd, NDF_ONLY_PNBUF);
435 		if (nd.ni_dvp == vp)
436 			vrele(nd.ni_dvp);
437 		else
438 			vput(nd.ni_dvp);
439 		if (vp != NULL) {
440 			vrele(vp);
441 			error = EADDRINUSE;
442 			goto error;
443 		}
444 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
445 		if (error)
446 			goto error;
447 		VFS_UNLOCK_GIANT(vfslocked);
448 		goto restart;
449 	}
450 	VATTR_NULL(&vattr);
451 	vattr.va_type = VSOCK;
452 	vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
453 #ifdef MAC
454 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
455 	    &vattr);
456 #endif
457 	if (error == 0) {
458 		VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
459 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
460 	}
461 	NDFREE(&nd, NDF_ONLY_PNBUF);
462 	vput(nd.ni_dvp);
463 	if (error) {
464 		vn_finished_write(mp);
465 		goto error;
466 	}
467 	vp = nd.ni_vp;
468 	ASSERT_VOP_ELOCKED(vp, "uipc_bind");
469 	soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
470 
471 	UNP_GLOBAL_WLOCK();
472 	UNP_PCB_LOCK(unp);
473 	vp->v_socket = unp->unp_socket;
474 	unp->unp_vnode = vp;
475 	unp->unp_addr = soun;
476 	unp->unp_flags &= ~UNP_BINDING;
477 	UNP_PCB_UNLOCK(unp);
478 	UNP_GLOBAL_WUNLOCK();
479 	VOP_UNLOCK(vp, 0);
480 	vn_finished_write(mp);
481 	VFS_UNLOCK_GIANT(vfslocked);
482 	free(buf, M_TEMP);
483 	return (0);
484 
485 error:
486 	VFS_UNLOCK_GIANT(vfslocked);
487 	UNP_PCB_LOCK(unp);
488 	unp->unp_flags &= ~UNP_BINDING;
489 	UNP_PCB_UNLOCK(unp);
490 	free(buf, M_TEMP);
491 	return (error);
492 }
493 
494 static int
495 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
496 {
497 	int error;
498 
499 	KASSERT(td == curthread, ("uipc_connect: td != curthread"));
500 	UNP_GLOBAL_WLOCK();
501 	error = unp_connect(so, nam, td);
502 	UNP_GLOBAL_WUNLOCK();
503 	return (error);
504 }
505 
506 static void
507 uipc_close(struct socket *so)
508 {
509 	struct unpcb *unp, *unp2;
510 
511 	unp = sotounpcb(so);
512 	KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
513 
514 	UNP_GLOBAL_WLOCK();
515 	UNP_PCB_LOCK(unp);
516 	unp2 = unp->unp_conn;
517 	if (unp2 != NULL) {
518 		UNP_PCB_LOCK(unp2);
519 		unp_disconnect(unp, unp2);
520 		UNP_PCB_UNLOCK(unp2);
521 	}
522 	UNP_PCB_UNLOCK(unp);
523 	UNP_GLOBAL_WUNLOCK();
524 }
525 
526 int
527 uipc_connect2(struct socket *so1, struct socket *so2)
528 {
529 	struct unpcb *unp, *unp2;
530 	int error;
531 
532 	UNP_GLOBAL_WLOCK();
533 	unp = so1->so_pcb;
534 	KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
535 	UNP_PCB_LOCK(unp);
536 	unp2 = so2->so_pcb;
537 	KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
538 	UNP_PCB_LOCK(unp2);
539 	error = unp_connect2(so1, so2, PRU_CONNECT2);
540 	UNP_PCB_UNLOCK(unp2);
541 	UNP_PCB_UNLOCK(unp);
542 	UNP_GLOBAL_WUNLOCK();
543 	return (error);
544 }
545 
546 /* control is EOPNOTSUPP */
547 
548 static void
549 uipc_detach(struct socket *so)
550 {
551 	struct unpcb *unp, *unp2;
552 	struct sockaddr_un *saved_unp_addr;
553 	struct vnode *vp;
554 	int freeunp, local_unp_rights;
555 
556 	unp = sotounpcb(so);
557 	KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
558 
559 	UNP_GLOBAL_WLOCK();
560 	UNP_PCB_LOCK(unp);
561 
562 	LIST_REMOVE(unp, unp_link);
563 	unp->unp_gencnt = ++unp_gencnt;
564 	--unp_count;
565 
566 	/*
567 	 * XXXRW: Should assert vp->v_socket == so.
568 	 */
569 	if ((vp = unp->unp_vnode) != NULL) {
570 		unp->unp_vnode->v_socket = NULL;
571 		unp->unp_vnode = NULL;
572 	}
573 	unp2 = unp->unp_conn;
574 	if (unp2 != NULL) {
575 		UNP_PCB_LOCK(unp2);
576 		unp_disconnect(unp, unp2);
577 		UNP_PCB_UNLOCK(unp2);
578 	}
579 
580 	/*
581 	 * We hold the global lock, so it's OK to acquire multiple pcb locks
582 	 * at a time.
583 	 */
584 	while (!LIST_EMPTY(&unp->unp_refs)) {
585 		struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
586 
587 		UNP_PCB_LOCK(ref);
588 		unp_drop(ref, ECONNRESET);
589 		UNP_PCB_UNLOCK(ref);
590 	}
591 	local_unp_rights = unp_rights;
592 	UNP_GLOBAL_WUNLOCK();
593 	unp->unp_socket->so_pcb = NULL;
594 	saved_unp_addr = unp->unp_addr;
595 	unp->unp_addr = NULL;
596 	unp->unp_refcount--;
597 	freeunp = (unp->unp_refcount == 0);
598 	if (saved_unp_addr != NULL)
599 		FREE(saved_unp_addr, M_SONAME);
600 	if (freeunp) {
601 		UNP_PCB_LOCK_DESTROY(unp);
602 		uma_zfree(unp_zone, unp);
603 	} else
604 		UNP_PCB_UNLOCK(unp);
605 	if (vp) {
606 		int vfslocked;
607 
608 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
609 		vrele(vp);
610 		VFS_UNLOCK_GIANT(vfslocked);
611 	}
612 	if (local_unp_rights)
613 		taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
614 }
615 
616 static int
617 uipc_disconnect(struct socket *so)
618 {
619 	struct unpcb *unp, *unp2;
620 
621 	unp = sotounpcb(so);
622 	KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
623 
624 	UNP_GLOBAL_WLOCK();
625 	UNP_PCB_LOCK(unp);
626 	unp2 = unp->unp_conn;
627 	if (unp2 != NULL) {
628 		UNP_PCB_LOCK(unp2);
629 		unp_disconnect(unp, unp2);
630 		UNP_PCB_UNLOCK(unp2);
631 	}
632 	UNP_PCB_UNLOCK(unp);
633 	UNP_GLOBAL_WUNLOCK();
634 	return (0);
635 }
636 
637 static int
638 uipc_listen(struct socket *so, int backlog, struct thread *td)
639 {
640 	struct unpcb *unp;
641 	int error;
642 
643 	unp = sotounpcb(so);
644 	KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
645 
646 	UNP_PCB_LOCK(unp);
647 	if (unp->unp_vnode == NULL) {
648 		UNP_PCB_UNLOCK(unp);
649 		return (EINVAL);
650 	}
651 
652 	SOCK_LOCK(so);
653 	error = solisten_proto_check(so);
654 	if (error == 0) {
655 		cru2x(td->td_ucred, &unp->unp_peercred);
656 		unp->unp_flags |= UNP_HAVEPCCACHED;
657 		solisten_proto(so, backlog);
658 	}
659 	SOCK_UNLOCK(so);
660 	UNP_PCB_UNLOCK(unp);
661 	return (error);
662 }
663 
664 static int
665 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
666 {
667 	struct unpcb *unp, *unp2;
668 	const struct sockaddr *sa;
669 
670 	unp = sotounpcb(so);
671 	KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
672 
673 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
674 	UNP_PCB_LOCK(unp);
675 	/*
676 	 * XXX: It seems that this test always fails even when connection is
677 	 * established.  So, this else clause is added as workaround to
678 	 * return PF_LOCAL sockaddr.
679 	 */
680 	unp2 = unp->unp_conn;
681 	if (unp2 != NULL) {
682 		UNP_PCB_LOCK(unp2);
683 		if (unp2->unp_addr != NULL)
684 			sa = (struct sockaddr *) unp->unp_conn->unp_addr;
685 		else
686 			sa = &sun_noname;
687 		bcopy(sa, *nam, sa->sa_len);
688 		UNP_PCB_UNLOCK(unp2);
689 	} else {
690 		sa = &sun_noname;
691 		bcopy(sa, *nam, sa->sa_len);
692 	}
693 	UNP_PCB_UNLOCK(unp);
694 	return (0);
695 }
696 
697 static int
698 uipc_rcvd(struct socket *so, int flags)
699 {
700 	struct unpcb *unp, *unp2;
701 	struct socket *so2;
702 	u_int mbcnt, sbcc;
703 	u_long newhiwat;
704 
705 	unp = sotounpcb(so);
706 	KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
707 
708 	if (so->so_type == SOCK_DGRAM)
709 		panic("uipc_rcvd DGRAM?");
710 
711 	if (so->so_type != SOCK_STREAM)
712 		panic("uipc_rcvd unknown socktype");
713 
714 	/*
715 	 * Adjust backpressure on sender and wakeup any waiting to write.
716 	 *
717 	 * The unp lock is acquired to maintain the validity of the unp_conn
718 	 * pointer; no lock on unp2 is required as unp2->unp_socket will be
719 	 * static as long as we don't permit unp2 to disconnect from unp,
720 	 * which is prevented by the lock on unp.  We cache values from
721 	 * so_rcv to avoid holding the so_rcv lock over the entire
722 	 * transaction on the remote so_snd.
723 	 */
724 	SOCKBUF_LOCK(&so->so_rcv);
725 	mbcnt = so->so_rcv.sb_mbcnt;
726 	sbcc = so->so_rcv.sb_cc;
727 	SOCKBUF_UNLOCK(&so->so_rcv);
728 	UNP_PCB_LOCK(unp);
729 	unp2 = unp->unp_conn;
730 	if (unp2 == NULL) {
731 		UNP_PCB_UNLOCK(unp);
732 		return (0);
733 	}
734 	so2 = unp2->unp_socket;
735 	SOCKBUF_LOCK(&so2->so_snd);
736 	so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
737 	newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
738 	(void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
739 	    newhiwat, RLIM_INFINITY);
740 	sowwakeup_locked(so2);
741 	unp->unp_mbcnt = mbcnt;
742 	unp->unp_cc = sbcc;
743 	UNP_PCB_UNLOCK(unp);
744 	return (0);
745 }
746 
747 /* pru_rcvoob is EOPNOTSUPP */
748 
749 static int
750 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
751     struct mbuf *control, struct thread *td)
752 {
753 	struct unpcb *unp, *unp2;
754 	struct socket *so2;
755 	u_int mbcnt, sbcc;
756 	u_long newhiwat;
757 	int error = 0;
758 
759 	unp = sotounpcb(so);
760 	KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
761 
762 	if (flags & PRUS_OOB) {
763 		error = EOPNOTSUPP;
764 		goto release;
765 	}
766 
767 	if (control != NULL && (error = unp_internalize(&control, td)))
768 		goto release;
769 
770 	if ((nam != NULL) || (flags & PRUS_EOF))
771 		UNP_GLOBAL_WLOCK();
772 	else
773 		UNP_GLOBAL_RLOCK();
774 
775 	switch (so->so_type) {
776 	case SOCK_DGRAM:
777 	{
778 		const struct sockaddr *from;
779 
780 		unp2 = unp->unp_conn;
781 		if (nam != NULL) {
782 			UNP_GLOBAL_WLOCK_ASSERT();
783 			if (unp2 != NULL) {
784 				error = EISCONN;
785 				break;
786 			}
787 			error = unp_connect(so, nam, td);
788 			if (error)
789 				break;
790 			unp2 = unp->unp_conn;
791 		}
792 		/*
793 		 * Because connect() and send() are non-atomic in a sendto()
794 		 * with a target address, it's possible that the socket will
795 		 * have disconnected before the send() can run.  In that case
796 		 * return the slightly counter-intuitive but otherwise
797 		 * correct error that the socket is not connected.
798 		 */
799 		if (unp2 == NULL) {
800 			error = ENOTCONN;
801 			break;
802 		}
803 		/* Lockless read. */
804 		if (unp2->unp_flags & UNP_WANTCRED)
805 			control = unp_addsockcred(td, control);
806 		UNP_PCB_LOCK(unp);
807 		if (unp->unp_addr != NULL)
808 			from = (struct sockaddr *)unp->unp_addr;
809 		else
810 			from = &sun_noname;
811 		so2 = unp2->unp_socket;
812 		SOCKBUF_LOCK(&so2->so_rcv);
813 		if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
814 			sorwakeup_locked(so2);
815 			m = NULL;
816 			control = NULL;
817 		} else {
818 			SOCKBUF_UNLOCK(&so2->so_rcv);
819 			error = ENOBUFS;
820 		}
821 		if (nam != NULL) {
822 			UNP_GLOBAL_WLOCK_ASSERT();
823 			UNP_PCB_LOCK(unp2);
824 			unp_disconnect(unp, unp2);
825 			UNP_PCB_UNLOCK(unp2);
826 		}
827 		UNP_PCB_UNLOCK(unp);
828 		break;
829 	}
830 
831 	case SOCK_STREAM:
832 		/*
833 		 * Connect if not connected yet.
834 		 *
835 		 * Note: A better implementation would complain if not equal
836 		 * to the peer's address.
837 		 */
838 		if ((so->so_state & SS_ISCONNECTED) == 0) {
839 			if (nam != NULL) {
840 				UNP_GLOBAL_WLOCK_ASSERT();
841 				error = unp_connect(so, nam, td);
842 				if (error)
843 					break;	/* XXX */
844 			} else {
845 				error = ENOTCONN;
846 				break;
847 			}
848 		}
849 
850 		/* Lockless read. */
851 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
852 			error = EPIPE;
853 			break;
854 		}
855 		/*
856 		 * Because connect() and send() are non-atomic in a sendto()
857 		 * with a target address, it's possible that the socket will
858 		 * have disconnected before the send() can run.  In that case
859 		 * return the slightly counter-intuitive but otherwise
860 		 * correct error that the socket is not connected.
861 		 *
862 		 * Locking here must be done carefully: the global lock
863 		 * prevents interconnections between unpcbs from changing, so
864 		 * we can traverse from unp to unp2 without acquiring unp's
865 		 * lock.  Socket buffer locks follow unpcb locks, so we can
866 		 * acquire both remote and lock socket buffer locks.
867 		 */
868 		unp2 = unp->unp_conn;
869 		if (unp2 == NULL) {
870 			error = ENOTCONN;
871 			break;
872 		}
873 		so2 = unp2->unp_socket;
874 		UNP_PCB_LOCK(unp2);
875 		SOCKBUF_LOCK(&so2->so_rcv);
876 		if (unp2->unp_flags & UNP_WANTCRED) {
877 			/*
878 			 * Credentials are passed only once on SOCK_STREAM.
879 			 */
880 			unp2->unp_flags &= ~UNP_WANTCRED;
881 			control = unp_addsockcred(td, control);
882 		}
883 		/*
884 		 * Send to paired receive port, and then reduce send buffer
885 		 * hiwater marks to maintain backpressure.  Wake up readers.
886 		 */
887 		if (control != NULL) {
888 			if (sbappendcontrol_locked(&so2->so_rcv, m, control))
889 				control = NULL;
890 		} else
891 			sbappend_locked(&so2->so_rcv, m);
892 		mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
893 		unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
894 		sbcc = so2->so_rcv.sb_cc;
895 		sorwakeup_locked(so2);
896 
897 		SOCKBUF_LOCK(&so->so_snd);
898 		newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
899 		(void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
900 		    newhiwat, RLIM_INFINITY);
901 		so->so_snd.sb_mbmax -= mbcnt;
902 		SOCKBUF_UNLOCK(&so->so_snd);
903 		unp2->unp_cc = sbcc;
904 		UNP_PCB_UNLOCK(unp2);
905 		m = NULL;
906 		break;
907 
908 	default:
909 		panic("uipc_send unknown socktype");
910 	}
911 
912 	/*
913 	 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
914 	 */
915 	if (flags & PRUS_EOF) {
916 		UNP_PCB_LOCK(unp);
917 		socantsendmore(so);
918 		unp_shutdown(unp);
919 		UNP_PCB_UNLOCK(unp);
920 	}
921 
922 	if ((nam != NULL) || (flags & PRUS_EOF))
923 		UNP_GLOBAL_WUNLOCK();
924 	else
925 		UNP_GLOBAL_RUNLOCK();
926 
927 	if (control != NULL && error != 0)
928 		unp_dispose(control);
929 
930 release:
931 	if (control != NULL)
932 		m_freem(control);
933 	if (m != NULL)
934 		m_freem(m);
935 	return (error);
936 }
937 
938 static int
939 uipc_sense(struct socket *so, struct stat *sb)
940 {
941 	struct unpcb *unp, *unp2;
942 	struct socket *so2;
943 
944 	unp = sotounpcb(so);
945 	KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
946 
947 	sb->st_blksize = so->so_snd.sb_hiwat;
948 	UNP_GLOBAL_RLOCK();
949 	UNP_PCB_LOCK(unp);
950 	unp2 = unp->unp_conn;
951 	if (so->so_type == SOCK_STREAM && unp2 != NULL) {
952 		so2 = unp2->unp_socket;
953 		sb->st_blksize += so2->so_rcv.sb_cc;
954 	}
955 	sb->st_dev = NODEV;
956 	if (unp->unp_ino == 0)
957 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
958 	sb->st_ino = unp->unp_ino;
959 	UNP_PCB_UNLOCK(unp);
960 	UNP_GLOBAL_RUNLOCK();
961 	return (0);
962 }
963 
964 static int
965 uipc_shutdown(struct socket *so)
966 {
967 	struct unpcb *unp;
968 
969 	unp = sotounpcb(so);
970 	KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
971 
972 	UNP_GLOBAL_WLOCK();
973 	UNP_PCB_LOCK(unp);
974 	socantsendmore(so);
975 	unp_shutdown(unp);
976 	UNP_PCB_UNLOCK(unp);
977 	UNP_GLOBAL_WUNLOCK();
978 	return (0);
979 }
980 
981 static int
982 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
983 {
984 	struct unpcb *unp;
985 	const struct sockaddr *sa;
986 
987 	unp = sotounpcb(so);
988 	KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
989 
990 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
991 	UNP_PCB_LOCK(unp);
992 	if (unp->unp_addr != NULL)
993 		sa = (struct sockaddr *) unp->unp_addr;
994 	else
995 		sa = &sun_noname;
996 	bcopy(sa, *nam, sa->sa_len);
997 	UNP_PCB_UNLOCK(unp);
998 	return (0);
999 }
1000 
1001 struct pr_usrreqs uipc_usrreqs = {
1002 	.pru_abort = 		uipc_abort,
1003 	.pru_accept =		uipc_accept,
1004 	.pru_attach =		uipc_attach,
1005 	.pru_bind =		uipc_bind,
1006 	.pru_connect =		uipc_connect,
1007 	.pru_connect2 =		uipc_connect2,
1008 	.pru_detach =		uipc_detach,
1009 	.pru_disconnect =	uipc_disconnect,
1010 	.pru_listen =		uipc_listen,
1011 	.pru_peeraddr =		uipc_peeraddr,
1012 	.pru_rcvd =		uipc_rcvd,
1013 	.pru_send =		uipc_send,
1014 	.pru_sense =		uipc_sense,
1015 	.pru_shutdown =		uipc_shutdown,
1016 	.pru_sockaddr =		uipc_sockaddr,
1017 	.pru_close =		uipc_close,
1018 };
1019 
1020 int
1021 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1022 {
1023 	struct unpcb *unp;
1024 	struct xucred xu;
1025 	int error, optval;
1026 
1027 	if (sopt->sopt_level != 0)
1028 		return (EINVAL);
1029 
1030 	unp = sotounpcb(so);
1031 	KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1032 	error = 0;
1033 	switch (sopt->sopt_dir) {
1034 	case SOPT_GET:
1035 		switch (sopt->sopt_name) {
1036 		case LOCAL_PEERCRED:
1037 			UNP_PCB_LOCK(unp);
1038 			if (unp->unp_flags & UNP_HAVEPC)
1039 				xu = unp->unp_peercred;
1040 			else {
1041 				if (so->so_type == SOCK_STREAM)
1042 					error = ENOTCONN;
1043 				else
1044 					error = EINVAL;
1045 			}
1046 			UNP_PCB_UNLOCK(unp);
1047 			if (error == 0)
1048 				error = sooptcopyout(sopt, &xu, sizeof(xu));
1049 			break;
1050 
1051 		case LOCAL_CREDS:
1052 			/* Unlocked read. */
1053 			optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1054 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1055 			break;
1056 
1057 		case LOCAL_CONNWAIT:
1058 			/* Unlocked read. */
1059 			optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1060 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1061 			break;
1062 
1063 		default:
1064 			error = EOPNOTSUPP;
1065 			break;
1066 		}
1067 		break;
1068 
1069 	case SOPT_SET:
1070 		switch (sopt->sopt_name) {
1071 		case LOCAL_CREDS:
1072 		case LOCAL_CONNWAIT:
1073 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1074 					    sizeof(optval));
1075 			if (error)
1076 				break;
1077 
1078 #define	OPTSET(bit) do {						\
1079 	UNP_PCB_LOCK(unp);						\
1080 	if (optval)							\
1081 		unp->unp_flags |= bit;					\
1082 	else								\
1083 		unp->unp_flags &= ~bit;					\
1084 	UNP_PCB_UNLOCK(unp);						\
1085 } while (0)
1086 
1087 			switch (sopt->sopt_name) {
1088 			case LOCAL_CREDS:
1089 				OPTSET(UNP_WANTCRED);
1090 				break;
1091 
1092 			case LOCAL_CONNWAIT:
1093 				OPTSET(UNP_CONNWAIT);
1094 				break;
1095 
1096 			default:
1097 				break;
1098 			}
1099 			break;
1100 #undef	OPTSET
1101 		default:
1102 			error = ENOPROTOOPT;
1103 			break;
1104 		}
1105 		break;
1106 
1107 	default:
1108 		error = EOPNOTSUPP;
1109 		break;
1110 	}
1111 	return (error);
1112 }
1113 
1114 static int
1115 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1116 {
1117 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1118 	struct vnode *vp;
1119 	struct socket *so2, *so3;
1120 	struct unpcb *unp, *unp2, *unp3;
1121 	int error, len, vfslocked;
1122 	struct nameidata nd;
1123 	char buf[SOCK_MAXADDRLEN];
1124 	struct sockaddr *sa;
1125 
1126 	UNP_GLOBAL_WLOCK_ASSERT();
1127 
1128 	unp = sotounpcb(so);
1129 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1130 
1131 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1132 	if (len <= 0)
1133 		return (EINVAL);
1134 	bcopy(soun->sun_path, buf, len);
1135 	buf[len] = 0;
1136 
1137 	UNP_PCB_LOCK(unp);
1138 	if (unp->unp_flags & UNP_CONNECTING) {
1139 		UNP_PCB_UNLOCK(unp);
1140 		return (EALREADY);
1141 	}
1142 	UNP_GLOBAL_WUNLOCK();
1143 	unp->unp_flags |= UNP_CONNECTING;
1144 	UNP_PCB_UNLOCK(unp);
1145 
1146 	sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1147 	NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf,
1148 	    td);
1149 	error = namei(&nd);
1150 	if (error)
1151 		vp = NULL;
1152 	else
1153 		vp = nd.ni_vp;
1154 	ASSERT_VOP_LOCKED(vp, "unp_connect");
1155 	vfslocked = NDHASGIANT(&nd);
1156 	NDFREE(&nd, NDF_ONLY_PNBUF);
1157 	if (error)
1158 		goto bad;
1159 
1160 	if (vp->v_type != VSOCK) {
1161 		error = ENOTSOCK;
1162 		goto bad;
1163 	}
1164 #ifdef MAC
1165 	error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1166 	if (error)
1167 		goto bad;
1168 #endif
1169 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1170 	if (error)
1171 		goto bad;
1172 	VFS_UNLOCK_GIANT(vfslocked);
1173 
1174 	unp = sotounpcb(so);
1175 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1176 
1177 	/*
1178 	 * Lock global lock for two reasons: make sure v_socket is stable,
1179 	 * and to protect simultaneous locking of multiple pcbs.
1180 	 */
1181 	UNP_GLOBAL_WLOCK();
1182 	so2 = vp->v_socket;
1183 	if (so2 == NULL) {
1184 		error = ECONNREFUSED;
1185 		goto bad2;
1186 	}
1187 	if (so->so_type != so2->so_type) {
1188 		error = EPROTOTYPE;
1189 		goto bad2;
1190 	}
1191 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1192 		if (so2->so_options & SO_ACCEPTCONN) {
1193 			/*
1194 			 * We can't drop the global lock here or 'so2' may
1195 			 * become invalid.  As a result, we need to handle
1196 			 * possibly lock recursion in uipc_attach.
1197 			 */
1198 			so3 = sonewconn(so2, 0);
1199 		} else
1200 			so3 = NULL;
1201 		if (so3 == NULL) {
1202 			error = ECONNREFUSED;
1203 			goto bad2;
1204 		}
1205 		unp = sotounpcb(so);
1206 		unp2 = sotounpcb(so2);
1207 		unp3 = sotounpcb(so3);
1208 		UNP_PCB_LOCK(unp);
1209 		UNP_PCB_LOCK(unp2);
1210 		UNP_PCB_LOCK(unp3);
1211 		if (unp2->unp_addr != NULL) {
1212 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1213 			unp3->unp_addr = (struct sockaddr_un *) sa;
1214 			sa = NULL;
1215 		}
1216 		/*
1217 		 * unp_peercred management:
1218 		 *
1219 		 * The connecter's (client's) credentials are copied from its
1220 		 * process structure at the time of connect() (which is now).
1221 		 */
1222 		cru2x(td->td_ucred, &unp3->unp_peercred);
1223 		unp3->unp_flags |= UNP_HAVEPC;
1224 		/*
1225 		 * The receiver's (server's) credentials are copied from the
1226 		 * unp_peercred member of socket on which the former called
1227 		 * listen(); uipc_listen() cached that process's credentials
1228 		 * at that time so we can use them now.
1229 		 */
1230 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1231 		    ("unp_connect: listener without cached peercred"));
1232 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1233 		    sizeof(unp->unp_peercred));
1234 		unp->unp_flags |= UNP_HAVEPC;
1235 		if (unp2->unp_flags & UNP_WANTCRED)
1236 			unp3->unp_flags |= UNP_WANTCRED;
1237 		UNP_PCB_UNLOCK(unp3);
1238 		UNP_PCB_UNLOCK(unp2);
1239 		UNP_PCB_UNLOCK(unp);
1240 #ifdef MAC
1241 		SOCK_LOCK(so);
1242 		mac_socketpeer_set_from_socket(so, so3);
1243 		mac_socketpeer_set_from_socket(so3, so);
1244 		SOCK_UNLOCK(so);
1245 #endif
1246 
1247 		so2 = so3;
1248 	}
1249 	unp = sotounpcb(so);
1250 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1251 	unp2 = sotounpcb(so2);
1252 	KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1253 	UNP_PCB_LOCK(unp);
1254 	UNP_PCB_LOCK(unp2);
1255 	error = unp_connect2(so, so2, PRU_CONNECT);
1256 	UNP_PCB_UNLOCK(unp2);
1257 	UNP_PCB_UNLOCK(unp);
1258 bad2:
1259 	UNP_GLOBAL_WUNLOCK();
1260 	if (vfslocked)
1261 		/*
1262 		 * Giant has been previously acquired. This means filesystem
1263 		 * isn't MPSAFE. Do it once again.
1264 		 */
1265 		mtx_lock(&Giant);
1266 bad:
1267 	if (vp != NULL)
1268 		vput(vp);
1269 	VFS_UNLOCK_GIANT(vfslocked);
1270 	free(sa, M_SONAME);
1271 	UNP_GLOBAL_WLOCK();
1272 	UNP_PCB_LOCK(unp);
1273 	unp->unp_flags &= ~UNP_CONNECTING;
1274 	UNP_PCB_UNLOCK(unp);
1275 	return (error);
1276 }
1277 
1278 static int
1279 unp_connect2(struct socket *so, struct socket *so2, int req)
1280 {
1281 	struct unpcb *unp;
1282 	struct unpcb *unp2;
1283 
1284 	unp = sotounpcb(so);
1285 	KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1286 	unp2 = sotounpcb(so2);
1287 	KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1288 
1289 	UNP_GLOBAL_WLOCK_ASSERT();
1290 	UNP_PCB_LOCK_ASSERT(unp);
1291 	UNP_PCB_LOCK_ASSERT(unp2);
1292 
1293 	if (so2->so_type != so->so_type)
1294 		return (EPROTOTYPE);
1295 	unp->unp_conn = unp2;
1296 
1297 	switch (so->so_type) {
1298 	case SOCK_DGRAM:
1299 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1300 		soisconnected(so);
1301 		break;
1302 
1303 	case SOCK_STREAM:
1304 		unp2->unp_conn = unp;
1305 		if (req == PRU_CONNECT &&
1306 		    ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1307 			soisconnecting(so);
1308 		else
1309 			soisconnected(so);
1310 		soisconnected(so2);
1311 		break;
1312 
1313 	default:
1314 		panic("unp_connect2");
1315 	}
1316 	return (0);
1317 }
1318 
1319 static void
1320 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1321 {
1322 	struct socket *so;
1323 
1324 	KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1325 
1326 	UNP_GLOBAL_WLOCK_ASSERT();
1327 	UNP_PCB_LOCK_ASSERT(unp);
1328 	UNP_PCB_LOCK_ASSERT(unp2);
1329 
1330 	unp->unp_conn = NULL;
1331 	switch (unp->unp_socket->so_type) {
1332 	case SOCK_DGRAM:
1333 		LIST_REMOVE(unp, unp_reflink);
1334 		so = unp->unp_socket;
1335 		SOCK_LOCK(so);
1336 		so->so_state &= ~SS_ISCONNECTED;
1337 		SOCK_UNLOCK(so);
1338 		break;
1339 
1340 	case SOCK_STREAM:
1341 		soisdisconnected(unp->unp_socket);
1342 		unp2->unp_conn = NULL;
1343 		soisdisconnected(unp2->unp_socket);
1344 		break;
1345 	}
1346 }
1347 
1348 /*
1349  * unp_pcblist() walks the global list of struct unpcb's to generate a
1350  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1351  * sequentially, validating the generation number on each to see if it has
1352  * been detached.  All of this is necessary because copyout() may sleep on
1353  * disk I/O.
1354  */
1355 static int
1356 unp_pcblist(SYSCTL_HANDLER_ARGS)
1357 {
1358 	int error, i, n;
1359 	int freeunp;
1360 	struct unpcb *unp, **unp_list;
1361 	unp_gen_t gencnt;
1362 	struct xunpgen *xug;
1363 	struct unp_head *head;
1364 	struct xunpcb *xu;
1365 
1366 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1367 
1368 	/*
1369 	 * The process of preparing the PCB list is too time-consuming and
1370 	 * resource-intensive to repeat twice on every request.
1371 	 */
1372 	if (req->oldptr == NULL) {
1373 		n = unp_count;
1374 		req->oldidx = 2 * (sizeof *xug)
1375 			+ (n + n/8) * sizeof(struct xunpcb);
1376 		return (0);
1377 	}
1378 
1379 	if (req->newptr != NULL)
1380 		return (EPERM);
1381 
1382 	/*
1383 	 * OK, now we're committed to doing something.
1384 	 */
1385 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1386 	UNP_GLOBAL_RLOCK();
1387 	gencnt = unp_gencnt;
1388 	n = unp_count;
1389 	UNP_GLOBAL_RUNLOCK();
1390 
1391 	xug->xug_len = sizeof *xug;
1392 	xug->xug_count = n;
1393 	xug->xug_gen = gencnt;
1394 	xug->xug_sogen = so_gencnt;
1395 	error = SYSCTL_OUT(req, xug, sizeof *xug);
1396 	if (error) {
1397 		free(xug, M_TEMP);
1398 		return (error);
1399 	}
1400 
1401 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1402 
1403 	UNP_GLOBAL_RLOCK();
1404 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1405 	     unp = LIST_NEXT(unp, unp_link)) {
1406 		UNP_PCB_LOCK(unp);
1407 		if (unp->unp_gencnt <= gencnt) {
1408 			if (cr_cansee(req->td->td_ucred,
1409 			    unp->unp_socket->so_cred)) {
1410 				UNP_PCB_UNLOCK(unp);
1411 				continue;
1412 			}
1413 			unp_list[i++] = unp;
1414 			unp->unp_refcount++;
1415 		}
1416 		UNP_PCB_UNLOCK(unp);
1417 	}
1418 	UNP_GLOBAL_RUNLOCK();
1419 	n = i;			/* In case we lost some during malloc. */
1420 
1421 	error = 0;
1422 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1423 	for (i = 0; i < n; i++) {
1424 		unp = unp_list[i];
1425 		UNP_PCB_LOCK(unp);
1426 		unp->unp_refcount--;
1427 	        if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1428 			xu->xu_len = sizeof *xu;
1429 			xu->xu_unpp = unp;
1430 			/*
1431 			 * XXX - need more locking here to protect against
1432 			 * connect/disconnect races for SMP.
1433 			 */
1434 			if (unp->unp_addr != NULL)
1435 				bcopy(unp->unp_addr, &xu->xu_addr,
1436 				      unp->unp_addr->sun_len);
1437 			if (unp->unp_conn != NULL &&
1438 			    unp->unp_conn->unp_addr != NULL)
1439 				bcopy(unp->unp_conn->unp_addr,
1440 				      &xu->xu_caddr,
1441 				      unp->unp_conn->unp_addr->sun_len);
1442 			bcopy(unp, &xu->xu_unp, sizeof *unp);
1443 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
1444 			UNP_PCB_UNLOCK(unp);
1445 			error = SYSCTL_OUT(req, xu, sizeof *xu);
1446 		} else {
1447 			freeunp = (unp->unp_refcount == 0);
1448 			UNP_PCB_UNLOCK(unp);
1449 			if (freeunp) {
1450 				UNP_PCB_LOCK_DESTROY(unp);
1451 				uma_zfree(unp_zone, unp);
1452 			}
1453 		}
1454 	}
1455 	free(xu, M_TEMP);
1456 	if (!error) {
1457 		/*
1458 		 * Give the user an updated idea of our state.  If the
1459 		 * generation differs from what we told her before, she knows
1460 		 * that something happened while we were processing this
1461 		 * request, and it might be necessary to retry.
1462 		 */
1463 		xug->xug_gen = unp_gencnt;
1464 		xug->xug_sogen = so_gencnt;
1465 		xug->xug_count = unp_count;
1466 		error = SYSCTL_OUT(req, xug, sizeof *xug);
1467 	}
1468 	free(unp_list, M_TEMP);
1469 	free(xug, M_TEMP);
1470 	return (error);
1471 }
1472 
1473 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1474 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1475 	    "List of active local datagram sockets");
1476 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1477 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1478 	    "List of active local stream sockets");
1479 
1480 static void
1481 unp_shutdown(struct unpcb *unp)
1482 {
1483 	struct unpcb *unp2;
1484 	struct socket *so;
1485 
1486 	UNP_GLOBAL_WLOCK_ASSERT();
1487 	UNP_PCB_LOCK_ASSERT(unp);
1488 
1489 	unp2 = unp->unp_conn;
1490 	if (unp->unp_socket->so_type == SOCK_STREAM && unp2 != NULL) {
1491 		so = unp2->unp_socket;
1492 		if (so != NULL)
1493 			socantrcvmore(so);
1494 	}
1495 }
1496 
1497 static void
1498 unp_drop(struct unpcb *unp, int errno)
1499 {
1500 	struct socket *so = unp->unp_socket;
1501 	struct unpcb *unp2;
1502 
1503 	UNP_GLOBAL_WLOCK_ASSERT();
1504 	UNP_PCB_LOCK_ASSERT(unp);
1505 
1506 	so->so_error = errno;
1507 	unp2 = unp->unp_conn;
1508 	if (unp2 == NULL)
1509 		return;
1510 
1511 	UNP_PCB_LOCK(unp2);
1512 	unp_disconnect(unp, unp2);
1513 	UNP_PCB_UNLOCK(unp2);
1514 }
1515 
1516 static void
1517 unp_freerights(struct file **rp, int fdcount)
1518 {
1519 	int i;
1520 	struct file *fp;
1521 
1522 	for (i = 0; i < fdcount; i++) {
1523 		/*
1524 		 * Zero the pointer before calling unp_discard since it may
1525 		 * end up in unp_gc()..
1526 		 *
1527 		 * XXXRW: This is less true than it used to be.
1528 		 */
1529 		fp = *rp;
1530 		*rp++ = NULL;
1531 		unp_discard(fp);
1532 	}
1533 }
1534 
1535 int
1536 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1537 {
1538 	struct thread *td = curthread;		/* XXX */
1539 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1540 	int i;
1541 	int *fdp;
1542 	struct file **rp;
1543 	struct file *fp;
1544 	void *data;
1545 	socklen_t clen = control->m_len, datalen;
1546 	int error, newfds;
1547 	int f;
1548 	u_int newlen;
1549 
1550 	UNP_GLOBAL_UNLOCK_ASSERT();
1551 
1552 	error = 0;
1553 	if (controlp != NULL) /* controlp == NULL => free control messages */
1554 		*controlp = NULL;
1555 
1556 	while (cm != NULL) {
1557 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1558 			error = EINVAL;
1559 			break;
1560 		}
1561 
1562 		data = CMSG_DATA(cm);
1563 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1564 
1565 		if (cm->cmsg_level == SOL_SOCKET
1566 		    && cm->cmsg_type == SCM_RIGHTS) {
1567 			newfds = datalen / sizeof(struct file *);
1568 			rp = data;
1569 
1570 			/* If we're not outputting the descriptors free them. */
1571 			if (error || controlp == NULL) {
1572 				unp_freerights(rp, newfds);
1573 				goto next;
1574 			}
1575 			FILEDESC_XLOCK(td->td_proc->p_fd);
1576 			/* if the new FD's will not fit free them.  */
1577 			if (!fdavail(td, newfds)) {
1578 				FILEDESC_XUNLOCK(td->td_proc->p_fd);
1579 				error = EMSGSIZE;
1580 				unp_freerights(rp, newfds);
1581 				goto next;
1582 			}
1583 			/*
1584 			 * Now change each pointer to an fd in the global
1585 			 * table to an integer that is the index to the local
1586 			 * fd table entry that we set up to point to the
1587 			 * global one we are transferring.
1588 			 */
1589 			newlen = newfds * sizeof(int);
1590 			*controlp = sbcreatecontrol(NULL, newlen,
1591 			    SCM_RIGHTS, SOL_SOCKET);
1592 			if (*controlp == NULL) {
1593 				FILEDESC_XUNLOCK(td->td_proc->p_fd);
1594 				error = E2BIG;
1595 				unp_freerights(rp, newfds);
1596 				goto next;
1597 			}
1598 
1599 			fdp = (int *)
1600 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1601 			for (i = 0; i < newfds; i++) {
1602 				if (fdalloc(td, 0, &f))
1603 					panic("unp_externalize fdalloc failed");
1604 				fp = *rp++;
1605 				td->td_proc->p_fd->fd_ofiles[f] = fp;
1606 				unp_externalize_fp(fp);
1607 				*fdp++ = f;
1608 			}
1609 			FILEDESC_XUNLOCK(td->td_proc->p_fd);
1610 		} else {
1611 			/* We can just copy anything else across. */
1612 			if (error || controlp == NULL)
1613 				goto next;
1614 			*controlp = sbcreatecontrol(NULL, datalen,
1615 			    cm->cmsg_type, cm->cmsg_level);
1616 			if (*controlp == NULL) {
1617 				error = ENOBUFS;
1618 				goto next;
1619 			}
1620 			bcopy(data,
1621 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1622 			    datalen);
1623 		}
1624 
1625 		controlp = &(*controlp)->m_next;
1626 
1627 next:
1628 		if (CMSG_SPACE(datalen) < clen) {
1629 			clen -= CMSG_SPACE(datalen);
1630 			cm = (struct cmsghdr *)
1631 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1632 		} else {
1633 			clen = 0;
1634 			cm = NULL;
1635 		}
1636 	}
1637 
1638 	m_freem(control);
1639 
1640 	return (error);
1641 }
1642 
1643 static void
1644 unp_zone_change(void *tag)
1645 {
1646 
1647 	uma_zone_set_max(unp_zone, maxsockets);
1648 }
1649 
1650 void
1651 unp_init(void)
1652 {
1653 
1654 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1655 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1656 	if (unp_zone == NULL)
1657 		panic("unp_init");
1658 	uma_zone_set_max(unp_zone, maxsockets);
1659 	EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1660 	    NULL, EVENTHANDLER_PRI_ANY);
1661 	LIST_INIT(&unp_dhead);
1662 	LIST_INIT(&unp_shead);
1663 	TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1664 	UNP_GLOBAL_LOCK_INIT();
1665 }
1666 
1667 static int
1668 unp_internalize(struct mbuf **controlp, struct thread *td)
1669 {
1670 	struct mbuf *control = *controlp;
1671 	struct proc *p = td->td_proc;
1672 	struct filedesc *fdescp = p->p_fd;
1673 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1674 	struct cmsgcred *cmcred;
1675 	struct file **rp;
1676 	struct file *fp;
1677 	struct timeval *tv;
1678 	int i, fd, *fdp;
1679 	void *data;
1680 	socklen_t clen = control->m_len, datalen;
1681 	int error, oldfds;
1682 	u_int newlen;
1683 
1684 	UNP_GLOBAL_UNLOCK_ASSERT();
1685 
1686 	error = 0;
1687 	*controlp = NULL;
1688 
1689 	while (cm != NULL) {
1690 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1691 		    || cm->cmsg_len > clen) {
1692 			error = EINVAL;
1693 			goto out;
1694 		}
1695 
1696 		data = CMSG_DATA(cm);
1697 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1698 
1699 		switch (cm->cmsg_type) {
1700 		/*
1701 		 * Fill in credential information.
1702 		 */
1703 		case SCM_CREDS:
1704 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1705 			    SCM_CREDS, SOL_SOCKET);
1706 			if (*controlp == NULL) {
1707 				error = ENOBUFS;
1708 				goto out;
1709 			}
1710 
1711 			cmcred = (struct cmsgcred *)
1712 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1713 			cmcred->cmcred_pid = p->p_pid;
1714 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1715 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1716 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1717 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1718 							CMGROUP_MAX);
1719 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
1720 				cmcred->cmcred_groups[i] =
1721 				    td->td_ucred->cr_groups[i];
1722 			break;
1723 
1724 		case SCM_RIGHTS:
1725 			oldfds = datalen / sizeof (int);
1726 			/*
1727 			 * Check that all the FDs passed in refer to legal
1728 			 * files.  If not, reject the entire operation.
1729 			 */
1730 			fdp = data;
1731 			FILEDESC_SLOCK(fdescp);
1732 			for (i = 0; i < oldfds; i++) {
1733 				fd = *fdp++;
1734 				if ((unsigned)fd >= fdescp->fd_nfiles ||
1735 				    fdescp->fd_ofiles[fd] == NULL) {
1736 					FILEDESC_SUNLOCK(fdescp);
1737 					error = EBADF;
1738 					goto out;
1739 				}
1740 				fp = fdescp->fd_ofiles[fd];
1741 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1742 					FILEDESC_SUNLOCK(fdescp);
1743 					error = EOPNOTSUPP;
1744 					goto out;
1745 				}
1746 
1747 			}
1748 
1749 			/*
1750 			 * Now replace the integer FDs with pointers to
1751 			 * the associated global file table entry..
1752 			 */
1753 			newlen = oldfds * sizeof(struct file *);
1754 			*controlp = sbcreatecontrol(NULL, newlen,
1755 			    SCM_RIGHTS, SOL_SOCKET);
1756 			if (*controlp == NULL) {
1757 				FILEDESC_SUNLOCK(fdescp);
1758 				error = E2BIG;
1759 				goto out;
1760 			}
1761 
1762 			fdp = data;
1763 			rp = (struct file **)
1764 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1765 			for (i = 0; i < oldfds; i++) {
1766 				fp = fdescp->fd_ofiles[*fdp++];
1767 				*rp++ = fp;
1768 				unp_internalize_fp(fp);
1769 			}
1770 			FILEDESC_SUNLOCK(fdescp);
1771 			break;
1772 
1773 		case SCM_TIMESTAMP:
1774 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
1775 			    SCM_TIMESTAMP, SOL_SOCKET);
1776 			if (*controlp == NULL) {
1777 				error = ENOBUFS;
1778 				goto out;
1779 			}
1780 			tv = (struct timeval *)
1781 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1782 			microtime(tv);
1783 			break;
1784 
1785 		default:
1786 			error = EINVAL;
1787 			goto out;
1788 		}
1789 
1790 		controlp = &(*controlp)->m_next;
1791 
1792 		if (CMSG_SPACE(datalen) < clen) {
1793 			clen -= CMSG_SPACE(datalen);
1794 			cm = (struct cmsghdr *)
1795 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1796 		} else {
1797 			clen = 0;
1798 			cm = NULL;
1799 		}
1800 	}
1801 
1802 out:
1803 	m_freem(control);
1804 
1805 	return (error);
1806 }
1807 
1808 static struct mbuf *
1809 unp_addsockcred(struct thread *td, struct mbuf *control)
1810 {
1811 	struct mbuf *m, *n, *n_prev;
1812 	struct sockcred *sc;
1813 	const struct cmsghdr *cm;
1814 	int ngroups;
1815 	int i;
1816 
1817 	ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1818 
1819 	m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1820 	if (m == NULL)
1821 		return (control);
1822 
1823 	sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1824 	sc->sc_uid = td->td_ucred->cr_ruid;
1825 	sc->sc_euid = td->td_ucred->cr_uid;
1826 	sc->sc_gid = td->td_ucred->cr_rgid;
1827 	sc->sc_egid = td->td_ucred->cr_gid;
1828 	sc->sc_ngroups = ngroups;
1829 	for (i = 0; i < sc->sc_ngroups; i++)
1830 		sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1831 
1832 	/*
1833 	 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1834 	 * created SCM_CREDS control message (struct sockcred) has another
1835 	 * format.
1836 	 */
1837 	if (control != NULL)
1838 		for (n = control, n_prev = NULL; n != NULL;) {
1839 			cm = mtod(n, struct cmsghdr *);
1840     			if (cm->cmsg_level == SOL_SOCKET &&
1841 			    cm->cmsg_type == SCM_CREDS) {
1842     				if (n_prev == NULL)
1843 					control = n->m_next;
1844 				else
1845 					n_prev->m_next = n->m_next;
1846 				n = m_free(n);
1847 			} else {
1848 				n_prev = n;
1849 				n = n->m_next;
1850 			}
1851 		}
1852 
1853 	/* Prepend it to the head. */
1854 	m->m_next = control;
1855 
1856 	return (m);
1857 }
1858 
1859 static struct unpcb *
1860 fptounp(struct file *fp)
1861 {
1862 	struct socket *so;
1863 
1864 	if (fp->f_type != DTYPE_SOCKET)
1865 		return (NULL);
1866 	if ((so = fp->f_data) == NULL)
1867 		return (NULL);
1868 	if (so->so_proto->pr_domain != &localdomain)
1869 		return (NULL);
1870 	return sotounpcb(so);
1871 }
1872 
1873 static void
1874 unp_discard(struct file *fp)
1875 {
1876 
1877 	unp_externalize_fp(fp);
1878 	(void) closef(fp, (struct thread *)NULL);
1879 }
1880 
1881 static void
1882 unp_internalize_fp(struct file *fp)
1883 {
1884 	struct unpcb *unp;
1885 
1886 	UNP_GLOBAL_WLOCK();
1887 	if ((unp = fptounp(fp)) != NULL) {
1888 		unp->unp_file = fp;
1889 		unp->unp_msgcount++;
1890 	}
1891 	fhold(fp);
1892 	unp_rights++;
1893 	UNP_GLOBAL_WUNLOCK();
1894 }
1895 
1896 static void
1897 unp_externalize_fp(struct file *fp)
1898 {
1899 	struct unpcb *unp;
1900 
1901 	UNP_GLOBAL_WLOCK();
1902 	if ((unp = fptounp(fp)) != NULL)
1903 		unp->unp_msgcount--;
1904 	unp_rights--;
1905 	UNP_GLOBAL_WUNLOCK();
1906 }
1907 
1908 /*
1909  * unp_defer indicates whether additional work has been defered for a future
1910  * pass through unp_gc().  It is thread local and does not require explicit
1911  * synchronization.
1912  */
1913 static int	unp_marked;
1914 static int	unp_unreachable;
1915 
1916 static void
1917 unp_accessable(struct file *fp)
1918 {
1919 	struct unpcb *unp;
1920 
1921 	if ((unp = fptounp(fp)) == NULL)
1922 		return;
1923 	if (unp->unp_gcflag & UNPGC_REF)
1924 		return;
1925 	unp->unp_gcflag &= ~UNPGC_DEAD;
1926 	unp->unp_gcflag |= UNPGC_REF;
1927 	unp_marked++;
1928 }
1929 
1930 static void
1931 unp_gc_process(struct unpcb *unp)
1932 {
1933 	struct socket *soa;
1934 	struct socket *so;
1935 	struct file *fp;
1936 
1937 	/* Already processed. */
1938 	if (unp->unp_gcflag & UNPGC_SCANNED)
1939 		return;
1940 	fp = unp->unp_file;
1941 	/*
1942 	 * Check for a socket potentially in a cycle.  It must be in a
1943 	 * queue as indicated by msgcount, and this must equal the file
1944 	 * reference count.  Note that when msgcount is 0 the file is NULL.
1945 	 */
1946 	if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
1947 	    unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
1948 		unp->unp_gcflag |= UNPGC_DEAD;
1949 		unp_unreachable++;
1950 		return;
1951 	}
1952 	/*
1953 	 * Mark all sockets we reference with RIGHTS.
1954 	 */
1955 	so = unp->unp_socket;
1956 	SOCKBUF_LOCK(&so->so_rcv);
1957 	unp_scan(so->so_rcv.sb_mb, unp_accessable);
1958 	SOCKBUF_UNLOCK(&so->so_rcv);
1959 	/*
1960 	 * Mark all sockets in our accept queue.
1961 	 */
1962 	ACCEPT_LOCK();
1963 	TAILQ_FOREACH(soa, &so->so_comp, so_list) {
1964 		SOCKBUF_LOCK(&soa->so_rcv);
1965 		unp_scan(soa->so_rcv.sb_mb, unp_accessable);
1966 		SOCKBUF_UNLOCK(&soa->so_rcv);
1967 	}
1968 	ACCEPT_UNLOCK();
1969 	unp->unp_gcflag |= UNPGC_SCANNED;
1970 }
1971 
1972 static int unp_recycled;
1973 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
1974     "Number of unreachable sockets claimed by the garbage collector.");
1975 
1976 static int unp_taskcount;
1977 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
1978     "Number of times the garbage collector has run.");
1979 
1980 static void
1981 unp_gc(__unused void *arg, int pending)
1982 {
1983 	struct unp_head *heads[] = { &unp_dhead, &unp_shead, NULL };
1984 	struct unp_head **head;
1985 	struct file **unref;
1986 	struct unpcb *unp;
1987 	int i;
1988 
1989 	unp_taskcount++;
1990 	UNP_GLOBAL_RLOCK();
1991 	/*
1992 	 * First clear all gc flags from previous runs.
1993 	 */
1994 	for (head = heads; *head != NULL; head++)
1995 		LIST_FOREACH(unp, *head, unp_link)
1996 			unp->unp_gcflag = 0;
1997 	/*
1998 	 * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
1999 	 * is reachable all of the sockets it references are reachable.
2000 	 * Stop the scan once we do a complete loop without discovering
2001 	 * a new reachable socket.
2002 	 */
2003 	do {
2004 		unp_unreachable = 0;
2005 		unp_marked = 0;
2006 		for (head = heads; *head != NULL; head++)
2007 			LIST_FOREACH(unp, *head, unp_link)
2008 				unp_gc_process(unp);
2009 	} while (unp_marked);
2010 	UNP_GLOBAL_RUNLOCK();
2011 	if (unp_unreachable == 0)
2012 		return;
2013 	/*
2014 	 * Allocate space for a local list of dead unpcbs.
2015 	 */
2016 	unref = malloc(unp_unreachable * sizeof(struct file *),
2017 	    M_TEMP, M_WAITOK);
2018 	/*
2019 	 * Iterate looking for sockets which have been specifically marked
2020 	 * as as unreachable and store them locally.
2021 	 */
2022 	UNP_GLOBAL_RLOCK();
2023 	for (i = 0, head = heads; *head != NULL; head++)
2024 		LIST_FOREACH(unp, *head, unp_link)
2025 			if (unp->unp_gcflag & UNPGC_DEAD) {
2026 				unref[i++] = unp->unp_file;
2027 				fhold(unp->unp_file);
2028 				KASSERT(unp->unp_file != NULL,
2029 				    ("unp_gc: Invalid unpcb."));
2030 				KASSERT(i <= unp_unreachable,
2031 				    ("unp_gc: incorrect unreachable count."));
2032 			}
2033 	UNP_GLOBAL_RUNLOCK();
2034 	/*
2035 	 * Now flush all sockets, free'ing rights.  This will free the
2036 	 * struct files associated with these sockets but leave each socket
2037 	 * with one remaining ref.
2038 	 */
2039 	for (i = 0; i < unp_unreachable; i++)
2040 		sorflush(unref[i]->f_data);
2041 	/*
2042 	 * And finally release the sockets so they can be reclaimed.
2043 	 */
2044 	for (i = 0; i < unp_unreachable; i++)
2045 		fdrop(unref[i], NULL);
2046 	unp_recycled += unp_unreachable;
2047 	free(unref, M_TEMP);
2048 }
2049 
2050 void
2051 unp_dispose(struct mbuf *m)
2052 {
2053 
2054 	if (m)
2055 		unp_scan(m, unp_discard);
2056 }
2057 
2058 static void
2059 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2060 {
2061 	struct mbuf *m;
2062 	struct file **rp;
2063 	struct cmsghdr *cm;
2064 	void *data;
2065 	int i;
2066 	socklen_t clen, datalen;
2067 	int qfds;
2068 
2069 	while (m0 != NULL) {
2070 		for (m = m0; m; m = m->m_next) {
2071 			if (m->m_type != MT_CONTROL)
2072 				continue;
2073 
2074 			cm = mtod(m, struct cmsghdr *);
2075 			clen = m->m_len;
2076 
2077 			while (cm != NULL) {
2078 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2079 					break;
2080 
2081 				data = CMSG_DATA(cm);
2082 				datalen = (caddr_t)cm + cm->cmsg_len
2083 				    - (caddr_t)data;
2084 
2085 				if (cm->cmsg_level == SOL_SOCKET &&
2086 				    cm->cmsg_type == SCM_RIGHTS) {
2087 					qfds = datalen / sizeof (struct file *);
2088 					rp = data;
2089 					for (i = 0; i < qfds; i++)
2090 						(*op)(*rp++);
2091 				}
2092 
2093 				if (CMSG_SPACE(datalen) < clen) {
2094 					clen -= CMSG_SPACE(datalen);
2095 					cm = (struct cmsghdr *)
2096 					    ((caddr_t)cm + CMSG_SPACE(datalen));
2097 				} else {
2098 					clen = 0;
2099 					cm = NULL;
2100 				}
2101 			}
2102 		}
2103 		m0 = m0->m_act;
2104 	}
2105 }
2106 
2107 #ifdef DDB
2108 static void
2109 db_print_indent(int indent)
2110 {
2111 	int i;
2112 
2113 	for (i = 0; i < indent; i++)
2114 		db_printf(" ");
2115 }
2116 
2117 static void
2118 db_print_unpflags(int unp_flags)
2119 {
2120 	int comma;
2121 
2122 	comma = 0;
2123 	if (unp_flags & UNP_HAVEPC) {
2124 		db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2125 		comma = 1;
2126 	}
2127 	if (unp_flags & UNP_HAVEPCCACHED) {
2128 		db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2129 		comma = 1;
2130 	}
2131 	if (unp_flags & UNP_WANTCRED) {
2132 		db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2133 		comma = 1;
2134 	}
2135 	if (unp_flags & UNP_CONNWAIT) {
2136 		db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2137 		comma = 1;
2138 	}
2139 	if (unp_flags & UNP_CONNECTING) {
2140 		db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2141 		comma = 1;
2142 	}
2143 	if (unp_flags & UNP_BINDING) {
2144 		db_printf("%sUNP_BINDING", comma ? ", " : "");
2145 		comma = 1;
2146 	}
2147 }
2148 
2149 static void
2150 db_print_xucred(int indent, struct xucred *xu)
2151 {
2152 	int comma, i;
2153 
2154 	db_print_indent(indent);
2155 	db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2156 	    xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2157 	db_print_indent(indent);
2158 	db_printf("cr_groups: ");
2159 	comma = 0;
2160 	for (i = 0; i < xu->cr_ngroups; i++) {
2161 		db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2162 		comma = 1;
2163 	}
2164 	db_printf("\n");
2165 }
2166 
2167 static void
2168 db_print_unprefs(int indent, struct unp_head *uh)
2169 {
2170 	struct unpcb *unp;
2171 	int counter;
2172 
2173 	counter = 0;
2174 	LIST_FOREACH(unp, uh, unp_reflink) {
2175 		if (counter % 4 == 0)
2176 			db_print_indent(indent);
2177 		db_printf("%p  ", unp);
2178 		if (counter % 4 == 3)
2179 			db_printf("\n");
2180 		counter++;
2181 	}
2182 	if (counter != 0 && counter % 4 != 0)
2183 		db_printf("\n");
2184 }
2185 
2186 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2187 {
2188 	struct unpcb *unp;
2189 
2190         if (!have_addr) {
2191                 db_printf("usage: show unpcb <addr>\n");
2192                 return;
2193         }
2194         unp = (struct unpcb *)addr;
2195 
2196 	db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2197 	    unp->unp_vnode);
2198 
2199 	db_printf("unp_ino: %d   unp_conn: %p\n", unp->unp_ino,
2200 	    unp->unp_conn);
2201 
2202 	db_printf("unp_refs:\n");
2203 	db_print_unprefs(2, &unp->unp_refs);
2204 
2205 	/* XXXRW: Would be nice to print the full address, if any. */
2206 	db_printf("unp_addr: %p\n", unp->unp_addr);
2207 
2208 	db_printf("unp_cc: %d   unp_mbcnt: %d   unp_gencnt: %llu\n",
2209 	    unp->unp_cc, unp->unp_mbcnt,
2210 	    (unsigned long long)unp->unp_gencnt);
2211 
2212 	db_printf("unp_flags: %x (", unp->unp_flags);
2213 	db_print_unpflags(unp->unp_flags);
2214 	db_printf(")\n");
2215 
2216 	db_printf("unp_peercred:\n");
2217 	db_print_xucred(2, &unp->unp_peercred);
2218 
2219 	db_printf("unp_refcount: %u\n", unp->unp_refcount);
2220 }
2221 #endif
2222